ParsNets: A Parsimonious Composition of Orthogonal and Low-Rank Linear Networks for Zero-Shot Learning

Abstract

We study the problem of design of strategy-proof in expectation (SP) mechanisms for facility location on a cycle, with the objective of minimizing the sum of costs of n agents. We show that there exists an SP mechanism that attains an approximation ratio of 7/4 with respect to the sum of costs of the agents, thus improving the best known upper bound of 2 - 2/n in the cases of n ≥ 5. The mechanism obtaining the bound randomizes between two mechanisms known in the literature: the Random Dictator (RD) and the Proportional Circle Distance (PCD) mechanism of Meir (2019). To prove the result, we propose a cycle-cutting technique that allows for estimating the problem on a cycle by a problem on a line.

Cite

Text

Guo et al. "ParsNets: A Parsimonious Composition of Orthogonal and Low-Rank Linear Networks for Zero-Shot Learning." International Joint Conference on Artificial Intelligence, 2024. doi:10.24963/ijcai.2024/449

Markdown

[Guo et al. "ParsNets: A Parsimonious Composition of Orthogonal and Low-Rank Linear Networks for Zero-Shot Learning." International Joint Conference on Artificial Intelligence, 2024.](https://mlanthology.org/ijcai/2024/guo2024ijcai-parsnets/) doi:10.24963/ijcai.2024/449

BibTeX

@inproceedings{guo2024ijcai-parsnets,
  title     = {{ParsNets: A Parsimonious Composition of Orthogonal and Low-Rank Linear Networks for Zero-Shot Learning}},
  author    = {Guo, Jingcai and Zhou, Qihua and Lu, Xiaocheng and Li, Ruibin and Liu, Ziming and Zhang, Jie and Han, Bo and Chen, Junyang and Xie, Xin and Guo, Song},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year      = {2024},
  pages     = {4062-4070},
  doi       = {10.24963/ijcai.2024/449},
  url       = {https://mlanthology.org/ijcai/2024/guo2024ijcai-parsnets/}
}